--- title: Title keywords: fastai sidebar: home_sidebar nb_path: "04e Embeddings of scRNA Velocity Data.ipynb" ---
{% raw %}
{% endraw %} {% raw %} 
from FRED.embed import ManifoldFlowEmbedder
from FRED.trainers import save_embedding_visualization, visualize_points
title = "Flow Neighbor Loss + Diffusion Map Reg"
MFE = ManifoldFlowEmbedder(
            embedding_dimension=2,
            embedder_shape=[4, 6, 8, 4, 2],
            device=device,
            sigma=0.5,
            flow_strength=0.5,
            smoothness_grid=True,
        )
loss_weights = {
            "reconstruction": 0,
            "diffusion map regularization": 1,
            "flow neighbor loss": 1,
            "smoothness": 0,
            "kld": 0,
        }
visualization_functions = [
    save_embedding_visualization,
    visualize_points
]
FREDtrainer = Trainer(FE = MFE, loss_weights=loss_weights, visualization_functions = visualization_functions, device=device, title = title)
{% endraw %} {% raw %} 
from FRED.datasets import double_helix, rnavelo
import scvelo as scv
from FRED.data_processing import dataloader_from_ndarray, ManifoldWithVectorField
from torch.utils.data import DataLoader
X, flow, labels = rnavelo(scv.datasets.simulation(n_obs=1000, switches=[0.2, 0.3, 0.5, 1]))

WARNING: Did not normalize X as it looks processed already. To enforce normalization, set `enforce=True`.
WARNING: Did not normalize spliced as it looks processed already. To enforce normalization, set `enforce=True`.
WARNING: Did not normalize unspliced as it looks processed already. To enforce normalization, set `enforce=True`.
Logarithmized X.
computing neighbors
    finished (0:00:00) --> added 
    'distances' and 'connectivities', weighted adjacency matrices (adata.obsp)
computing moments based on connectivities
    finished (0:00:00) --> added 
    'Ms' and 'Mu', moments of un/spliced abundances (adata.layers)
computing velocities
WARNING: Too few genes are selected as velocity genes. Consider setting a lower threshold for min_r2 or min_likelihood.
    finished (0:00:00) --> added 
    'velocity', velocity vectors for each individual cell (adata.layers)
{% endraw %} {% raw %} 
flow.shape

torch.Size([1000, 4])
{% endraw %} {% raw %} 
ds = ManifoldWithVectorField(X, flow, labels, sigma=1, flow_strength=5, dmap_coords_to_use=4, nbhd_strategy="flow neighbors", n_neighbors=5)
dataloader = DataLoader(ds, batch_size=None, shuffle=True)
{% endraw %} {% raw %} 
DC = dataloader.dataset.diff_coords
plt.scatter(DC[:,1],DC[:,2],c=labels)

<matplotlib.collections.PathCollection at 0x2a5c18490>
{% endraw %} {% raw %} 
FREDtrainer.fit(dataloader, n_epochs = 100)
{% endraw %} {% raw %} 
FREDtrainer.visualize_loss()
{% endraw %} {% raw %} 
FREDtrainer.training_gif(duration=150)
{% endraw %}